[Core] Multi-tenancy: enable multi-tenancy by default (#10570)

* Add new job in Travis to enable multi-tenancy

* fix

* Update .bazelrc

* Update .travis.yml

* fix test_job_gc_with_detached_actor

* fix test_multiple_downstream_tasks

* fix lint

* Enable multi-tenancy by default

* Kill idle workers in FIFO order

* Update test

* minor update

* Address comments

* fix some cases

* fix test_remote_cancel

* Address comments

* fix after merge

* remove kill

* fix worker_pool_test

* fix java test timeout

* fix test_two_custom_resources

* Add a delay when killing idle workers

* fix test_worker_failure

* fix test_worker_failed again

* fix DisconnectWorker

* update test_worker_failed

* Revert some python tests

* lint

* address comments
This commit is contained in:
Kai Yang
2020-09-29 23:54:53 -07:00
committed by GitHub
parent f54f7b23f6
commit 3504391fd2
17 changed files with 205 additions and 72 deletions
+15 -1
View File
@@ -11,7 +11,10 @@ import ray
import ray.cluster_utils
import ray.test_utils
from ray.test_utils import RayTestTimeoutException
from ray.test_utils import (
RayTestTimeoutException,
wait_for_condition,
)
logger = logging.getLogger(__name__)
@@ -505,6 +508,17 @@ def test_two_custom_resources(ray_start_cluster):
})
ray.init(address=cluster.address)
@ray.remote
def foo():
# Sleep a while to emulate a slow operation. This is needed to make
# sure tasks are scheduled to different nodes.
time.sleep(0.1)
return ray.worker.global_worker.node.unique_id
# Make sure each node has at least one idle worker.
wait_for_condition(
lambda: len(set(ray.get([foo.remote() for _ in range(6)]))) == 2)
@ray.remote(resources={"CustomResource1": 1})
def f():
time.sleep(0.001)